[Intel-gfx] [PATCH v6 10/64] drm/i915: make lockdep slightly happier about execbuf.

[Maarten Lankhorst]

As soon as we install fences, we should stop allocating memory
in order to prevent any potential deadlocks.

This is required later on, when we start adding support for
dma-fence annotations.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom at linux.intel.com>
---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 24 ++++++++++++++-----
 drivers/gpu/drm/i915/i915_active.c            | 20 ++++++++--------
 drivers/gpu/drm/i915/i915_vma.c               |  8 ++++---
 drivers/gpu/drm/i915/i915_vma.h               |  3 +++
 4 files changed, 36 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index d5666157dfa6..9c7480119d53 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -50,11 +50,12 @@ enum {
 #define DBG_FORCE_RELOC 0 /* choose one of the above! */
 };
 
-#define __EXEC_OBJECT_HAS_PIN		BIT(31)
-#define __EXEC_OBJECT_HAS_FENCE		BIT(30)
-#define __EXEC_OBJECT_NEEDS_MAP		BIT(29)
-#define __EXEC_OBJECT_NEEDS_BIAS	BIT(28)
-#define __EXEC_OBJECT_INTERNAL_FLAGS	(~0u << 28) /* all of the above */
+/* __EXEC_OBJECT_NO_RESERVE is BIT(31), defined in i915_vma.h */
+#define __EXEC_OBJECT_HAS_PIN		BIT(30)
+#define __EXEC_OBJECT_HAS_FENCE		BIT(29)
+#define __EXEC_OBJECT_NEEDS_MAP		BIT(28)
+#define __EXEC_OBJECT_NEEDS_BIAS	BIT(27)
+#define __EXEC_OBJECT_INTERNAL_FLAGS	(~0u << 27) /* all of the above + */
 #define __EXEC_OBJECT_RESERVED (__EXEC_OBJECT_HAS_PIN | __EXEC_OBJECT_HAS_FENCE)
 
 #define __EXEC_HAS_RELOC	BIT(31)
@@ -928,6 +929,12 @@ static int eb_validate_vmas(struct i915_execbuffer *eb)
 			}
 		}
 
+		if (!(ev->flags & EXEC_OBJECT_WRITE)) {
+			err = dma_resv_reserve_shared(vma->resv, 1);
+			if (err)
+				return err;
+		}
+
 		GEM_BUG_ON(drm_mm_node_allocated(&vma->node) &&
 			   eb_vma_misplaced(&eb->exec[i], vma, ev->flags));
 	}
@@ -2195,7 +2202,8 @@ static int eb_move_to_gpu(struct i915_execbuffer *eb)
 		}
 
 		if (err == 0)
-			err = i915_vma_move_to_active(vma, eb->request, flags);
+			err = i915_vma_move_to_active(vma, eb->request,
+						      flags | __EXEC_OBJECT_NO_RESERVE);
 	}
 
 	if (unlikely(err))
@@ -2447,6 +2455,10 @@ static int eb_parse_pipeline(struct i915_execbuffer *eb,
 	if (err)
 		goto err_commit;
 
+	err = dma_resv_reserve_shared(shadow->resv, 1);
+	if (err)
+		goto err_commit;
+
 	/* Wait for all writes (and relocs) into the batch to complete */
 	err = i915_sw_fence_await_reservation(&pw->base.chain,
 					      pw->batch->resv, NULL, false,
diff --git a/drivers/gpu/drm/i915/i915_active.c b/drivers/gpu/drm/i915/i915_active.c
index ab4382841c6b..4e78de64b71b 100644
--- a/drivers/gpu/drm/i915/i915_active.c
+++ b/drivers/gpu/drm/i915/i915_active.c
@@ -293,18 +293,13 @@ static struct active_node *__active_lookup(struct i915_active *ref, u64 idx)
 static struct i915_active_fence *
 active_instance(struct i915_active *ref, u64 idx)
 {
-	struct active_node *node, *prealloc;
+	struct active_node *node;
 	struct rb_node **p, *parent;
 
 	node = __active_lookup(ref, idx);
 	if (likely(node))
 		return &node->base;
 
-	/* Preallocate a replacement, just in case */
-	prealloc = kmem_cache_alloc(global.slab_cache, GFP_KERNEL);
-	if (!prealloc)
-		return NULL;
-
 	spin_lock_irq(&ref->tree_lock);
 	GEM_BUG_ON(i915_active_is_idle(ref));
 
@@ -314,10 +309,8 @@ active_instance(struct i915_active *ref, u64 idx)
 		parent = *p;
 
 		node = rb_entry(parent, struct active_node, node);
-		if (node->timeline == idx) {
-			kmem_cache_free(global.slab_cache, prealloc);
+		if (node->timeline == idx)
 			goto out;
-		}
 
 		if (node->timeline < idx)
 			p = &parent->rb_right;
@@ -325,7 +318,14 @@ active_instance(struct i915_active *ref, u64 idx)
 			p = &parent->rb_left;
 	}
 
-	node = prealloc;
+	/*
+	 * XXX: We should preallocate this before i915_active_ref() is ever
+	 *  called, but we cannot call into fs_reclaim() anyway, so use GFP_ATOMIC.
+	 */
+	node = kmem_cache_alloc(global.slab_cache, GFP_ATOMIC);
+	if (!node)
+		goto out;
+
 	__i915_active_fence_init(&node->base, NULL, node_retire);
 	node->ref = ref;
 	node->timeline = idx;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 7310893086f7..1ffda2aaa7a0 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1247,9 +1247,11 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 		obj->write_domain = I915_GEM_DOMAIN_RENDER;
 		obj->read_domains = 0;
 	} else {
-		err = dma_resv_reserve_shared(vma->resv, 1);
-		if (unlikely(err))
-			return err;
+		if (!(flags & __EXEC_OBJECT_NO_RESERVE)) {
+			err = dma_resv_reserve_shared(vma->resv, 1);
+			if (unlikely(err))
+				return err;
+		}
 
 		dma_resv_add_shared_fence(vma->resv, &rq->fence);
 		obj->write_domain = 0;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 838bbbeb11cc..3c951d5428cf 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -52,6 +52,9 @@ static inline bool i915_vma_is_active(const struct i915_vma *vma)
 	return !i915_active_is_idle(&vma->active);
 }
 
+/* do not reserve memory to prevent deadlocks */
+#define __EXEC_OBJECT_NO_RESERVE BIT(31)
+
 int __must_check __i915_vma_move_to_active(struct i915_vma *vma,
 					   struct i915_request *rq);
 int __must_check i915_vma_move_to_active(struct i915_vma *vma,
-- 
2.30.0.rc1